Query a data lake or lakehouse by using Azure Synapse serverless

Data Lake
Data Lake Storage
Data Lake Storage Gen2
Synapse Analytics
Blob Storage

This article describes an alternative approach to data warehouse projects that's called exploratory data analysis (EDA). This approach can reduce the challenges of extract, transform, load (ETL) operations. It focuses first on generating business insights and then turns to solving the modeling and ETL tasks.

Architecture

Diagram that shows a sample EDA architecture.

Download a Visio file of this architecture.

For EDA, you're concerned only with the right side of the diagram. Azure Synapse SQL serverless is used as the compute engine over the data lake files.

To accomplish EDA:

  • T-SQL queries run directly in Azure Synapse SQL serverless or Azure Synapse Spark.
  • Queries run from a graphical query tool like Power BI or Azure Data Studio.

We recommend that you persist all lakehouse data by using Parquet or Delta.

You can implement the left side of the diagram (data ingestion) by using any extract, load, transform (ELT) tool. It has no effect on EDA.

Components

Alternatives

  • You can replace or complement Synapse SQL serverless pools with Azure Databricks.

  • Instead of using a lakehouse model with Synapse SQL serverless pools, you can use Azure Synapse dedicated SQL pools to store enterprise data. Review the use cases and considerations in this article and related resources to decide which technology to use.

Scenario details

This solution shows an implementation of the EDA approach to data warehouse projects. This approach can reduce the challenges of ETL operations. It focuses first on generating business insights and then turns to solving modeling and ETL tasks.

Potential use cases

Other scenarios that can benefit from this analytics pattern:

  • Prescriptive analytics. Ask questions of your data, like Next Best Action, or what do we do next? Use data to be more data-driven and less gut-driven. The data might be unstructured and from many external sources of varying quality. You might want to use the data as fast as possible to evaluate your business strategy without actually loading the data into a data warehouse. You might dispose of the data after you answer your questions.

  • Self-service ETL. Do ETL/ELT when you do your data sandboxing (EDA) activities. Transform data and make it valuable. Doing so can improve the scale of your ETL developers.

About exploratory data analysis

Before we look more closely at how EDA works, it's worth summarizing the traditional approach to data warehouse projects. The traditional approach looks like this:

  • Requirements gathering. Document what to do with the data.

  • Data modeling. Determine how to model the numeric and attribute data into fact and dimension tables. Traditionally, you do this step before you acquire the new data.

  • ETL. Acquire the data and massage it into the data warehouse's data model.

These steps can take weeks or even months. Only then can you begin to query the data and solve the business problem. The user sees value only after the reports are created. The solution architecture usually looks something like this one:

Diagram that shows the traditional data warehouse architecture.

You can do this in another way that focuses first on generating business insights and then turns to solving the modeling and ETL tasks. The process is similar to data science processes. It looks something like this:

Diagram that describes data sandboxing.

In the industry, this process is called EDA, or exploratory data analysis.

Here are the steps:

  • Data acquisition. First, you need to determine what data sources you need to ingest into your data lake / sandbox. You then need to bring that data into the landing area of your lake. Azure provides tools like Azure Data Factory and Azure Logic Apps that can ingest data quickly.

  • Data sandboxing. Initially, a business analyst and an engineer who's skilled in exploratory data analysis via Azure Synapse Analytics serverless or basic SQL work together. During this phase, they're trying to uncover the business insight by using the new data. EDA is an iterative process. You might need to ingest more data, talk with SMEs, ask more questions, or generate visualizations.

  • Evaluation. After you find the business insight, you need to evaluate what to do with the data. You might want to persist the data into the data warehouse (so you move to the modeling phase). In other cases, you might decide to keep the data in the data lake / lakehouse and use it for predictive analytics (machine learning algorithms). In still other cases, you might decide to backfill your systems of record with the new insights. Based on these decisions, you can gain a better understanding of what you need to do next. You might not need to do ETL.

These methods are the core of true self-service analytics. By using the data lake and a query tool like Azure Synapse serverless that understands data lake query patterns, you can put your data assets into the hands of business people who understand a modicum of SQL. You can radically shorten the time-to-value by using this method and remove some of the risk associated with corporate data initiatives.

Considerations

These considerations implement the pillars of the Azure Well-Architected Framework, which is a set of guiding tenets that can be used to improve the quality of a workload. For more information, see Microsoft Azure Well-Architected Framework.

Availability

Azure Synapse SQL serverless pools is a platform as a service (PaaS) feature that can meet your high availability (HA) and disaster recovery (DR) requirements.

Serverless pools are available on-demand. They don't require scaling up, down, in, or out or administration of any kind. They use a pay-per-query model, so there's no unused capacity at any time. Serverless pools are ideal for:

  • Ad-hoc data-science explorations in T-SQL.
  • Early prototyping for data warehouse entities.
  • Defining views that consumers can use, for example in Power BI, for scenarios that can tolerate performance lag.
  • Exploratory data analysis.

Operations

Synapse SQL serverless uses standard T-SQL for querying and operations. You could use Synapse workspace UI, Azure Data Studio, or SQL Server Management Studio as the T-SQL tool.

Cost optimization

Cost optimization is about looking at ways to reduce unnecessary expenses and improve operational efficiencies. For more information, see Overview of the cost optimization pillar.

  • Data Lake Storage pricing depends on the amount of data you store and how often you use the data. The sample pricing includes one TB of data stored, with further transactional assumptions. The one TB refers to the size of the data lake, not the size of the original legacy database.

  • Azure Synapse Spark pool bases pricing on node size, number of instances, and uptime. The example assumes one small compute node with utilization between five hours per week and 40 hours per month.

  • Azure Synapse serverless SQL pool bases pricing on TBs of data processed. The sample assumes 50 TBs processed per month. This figure refers to the size of the data lake, not the size of the original legacy database.

Contributors

This article is being updated and maintained by Microsoft. It was originally written by the following contributors.

Principal authors:

Next steps